Researchers from Apple and Cornell have introduced STARFlow-V, a novel video generation model utilizing normalizing flows. This approach offers an alternative to diffusion models, achieving comparable visual quality while providing superior likelihood estimation and causal prediction capabilities. The model was presented at ICLR 2026. AI
Summary written by gemini-2.5-flash-lite from 3 sources. How we write summaries →
IMPACT Presents a new architectural approach to video generation, potentially challenging the dominance of diffusion models.
RANK_REASON Introduces a new model architecture for video generation presented at an academic conference.
